| |
Abstract:
Most generative grammarians (but see Sobin, 1987) assume that
English grammar is subject to the complementizer-trace (c-t)
constraint, which states that the dependency across a
complementizer (e.g., that) between a wh-filler and a gap in
subject position is ungrammatical (see (1)). Yet it has repeatedly
been observed that native speakers' judgments on violations of the
c-t constraint are not as clear-cut as on those ofother movement
constraints (cf. Bley-Vroman et al., 1988).
(1) *Who did you say that __ kissed Mary?
This study started as an attempt to use the logic of the
filled-gap effect (Stowe, 1986) to probe further English speakers'
knowledge of the c-t constraint and to test the role of the c-t
constraint in sentence processing. It assumes the gap-filling
metaphor of the processing of the wh-construction. 24 subjects read
28 sentences like those in (2) in the self-paced word-by-word
reading paradigm.
(2a) The tabloids do not report what flowers the proud mother
announced that the prince had bought __ for her daughter.
(2b) The tabloids do not report whether the proud mother announced
that the prince had bought roses for her daughter.
(2c) The tabloids do not report what flowers the proud mother
announced the prince had bought __ for her daughter.
(2d) The tabloids do not report whether the proud mother announced
the prince had bought roses for her daughter.
The (original) reasoning of the experiment is as follows: In (2a)
and (2c), the parser should start searching for a gap after reading
`what flowers'. If it expects a subject gap after `announced'
(having noted the implausibility of the wh-NP as the object of the
verb), a filled-gap effect should be found at `the prince' in (2c)
relative to (2d). (The wh-NP/verb combinations used in the
experiment were adapted from Garnsey et al. (1997); they each
involved an implausible wh-NP and a verb that does not prefer a
direct object to a sentential complement.) If the parser does not
follow the c-t constraint, a similar filled-gap effect ought to be
found at `the prince' in (2a) relative to (2b); but if the parser
is sensitive to the constraint, `that' should halt the expectation
of a subject gap and hence prevent a filled-gap effect in (2a) and
(2b). The following reading times for `the prince' (in msecs. per
word) were found: (2a) 361; (2b) 360; (2c) 399; (2d) 362. This
significant interaction, however, cannot straightforwardly be taken
as evidence for the parser's sensitivity to the c-t constraint
because, unexpectedly, reading time data for the rest of the
sentence offer conflicting information about the validity of the
interpretation of the filled-gap effect in (2c) and (2d) as the
parser's expectation of a subject gap rather than an object gap:
for, lack of filled-gap effect on `that' (and the use of
implausible wh-NP/verb combinations); against, non-significantly
longer reading time for `had bought' in (2c) than in (2a).
The lack of a filled-gap effect in (2a) and (2b) suggests that the
parser's often zealous gap-finding routine is unlikely to be the
source of native speakers' variable judgments of c-t violations. It
remains to be seen whether the null effect is due to knowledge of
the c-t constraint per se or is another manifestation of the
related observation that the parser does not expect a subject gap
right next to a wh-filler (cf. Gibson et al., 1994).
|
